Privacy notice

Last updated: 29 September 2026.

This notice describes the Flowstate beta at flowstate.icey.se. Flowstate provides visual project planning, collaboration, file storage, public project links, and optional AI access grants.

Data Flowstate processes

Project walkthrough recording starts only after the browser's capture permission picker; microphone narration is optional and separately permissioned. Recording bytes stay temporarily in the recording tab's memory until you choose Add to Board or Download. Add uploads an ordinary private Video with normal Project access, quota and retention. Download saves a file on your device that you may share yourself. Flowstate does not automatically upload an ongoing capture.

Owner-entered Project connection configurations contain server-encrypted service credentials and non-secret names, types and permission settings. Credential values are never read back into the form, AI context, Project snapshots, copies or exports. They remain encrypted Project data in the database and its backups; deleting a connection removes its live configuration, and permanent Project deletion removes its connections. Editing replaces credentials rather than revealing saved values.

Owner-entered Project secrets (for example a password or API token) are stored encrypted with a name and description, are never shown again in Flowstate, and are left out of Project snapshots, history, copies and exports. Only when the owner marks a secret "AI may read" can an external AI agent that the owner gave Edit access read its value through Flowstate's MCP tool; that agent's provider then holds the value under its own terms. Each such read is recorded with the time and the AI access used, and is shown to the owner. Deleting a secret removes it and its read history; permanent Project deletion removes all of them.

When the owner enables an action and grants Edit AI access, Flowstate may contact that configured service on the owner's behalf. The remote service receives its required authentication and bounded request; the authorized AI receives the permitted result, never the saved credential. External results may contain private service data. Known credential values are redacted, completed responses are encrypted for replay, and audits record identities/actions only. Up to 1,000 calls per connection are retained until that connection or Project is permanently deleted; ordinary database backups retain their existing policy. Disabling an action or revoking AI access stops subsequent use, including cached replay.

GIF animation is delivered from a sanitized version of the retained private original. Raster frames, timing and looping are kept; comments, text/application metadata and trailing content are removed. GIF delivery uses Flowstate's authorized same-origin proxy even when private storage is on S3. This restores earlier uploads without making original files public. Sanitized presentation bytes may be cached privately on the server for up to 24 hours within a 128 MiB cache budget. Authorization is checked before every delivery; deletion invalidates cached copies and maintenance prunes expired entries.

Account data: email address, display name, optional first and last name, password hash, verification state, assigned plan, and account status.
Project data: projects, boards, objects, blocks, connections, references, tasks, revisions, instructions, memberships, invitations, and activity needed to operate collaboration.
Uploaded content: private images (including images selected as object backgrounds), documents, video, and audio plus technical metadata such as filename, MIME type, byte size, checksum, storage location, and whether an object uses an Asset as content or background.
Security and operations data: sessions, optional remember-me tokens, password-reset and verification records, rate-limit records, audit events, IP address, user agent, application logs, maintenance state, and delivery status for configured email or project integrations.
Optional sharing data: revocable public-link capabilities, time-limited AI grants, AI project conversations, External AI Agent Bridge connections and transcripts, and integration settings created by a project owner.
Optional External Media data: a Project may store a validated YouTube or direct HTTPS Video/Audio reference. Flowstate does not copy those media bytes into its storage or count them toward quota.
Optional Local AI data: device-local tier/recommendation/failure preferences, installation metadata, and several gigabytes of explicitly downloaded model/tokenizer/WebAssembly artifacts in browser storage. The device benchmark stays in the browser.
Optional Local AI sources: TXT, Markdown, JSON, CSV, and HTML attached from the device are read and bounded in the browser without upload. If the user chooses URL or Data source instead, Flowstate servers fetch the public textual URL or authorized Project integration/cache and return bounded provenance- labeled content. Integration credentials remain server-side. Current Local AI does not extract PDF/DOCX text or inspect image pixels.

Why data is processed

Data is used to create and secure accounts, provide requested project and collaboration features, store and serve private content, enforce roles and quotas, recover from failures, prevent abuse, and meet requests such as export or deletion. Flowstate does not currently run advertising or behavioral analytics.

Flowstate uses server infrastructure, a private S3-compatible Project Asset store, a physically separate public S3-compatible distribution store, database storage, and outbound email infrastructure to operate the service. Local AI downloads exact immutable model artifacts directly from the Flowstate-controlled distribution store, but inference runs on the device and sends no prompt, Project context, answer, or benchmark to a third-party AI provider. Normal Project data, AI Chat messages, server-authorized mutations, and operational records still use Flowstate servers. Project owners can deliberately configure public links, AI grants, invitations, webhooks, or external data integrations; those features disclose only the scoped information their owner-authorized contract describes. No analytics provider or third-party browser script is currently configured.

When Local AI uses an attached local text file, its supported content is passed only to the on-device model and normal Chat persistence does not store the file body as a separate uploaded Asset. When the user explicitly attaches a URL or Data source, Flowstate servers process that retrieval and the resulting content is then supplied to the local model; no external AI provider receives it.

An owner-configured External AI Agent Bridge connection may send signed HTTPS callback events to the provider/runner selected by that owner. Callback payloads carry scoped identifiers, not complete credentials, session values, bearer tokens, private attachment URLs, cookies, or CSRF values. The external provider then processes only content it retrieves through the connection's current Listener/Responder and Read/Edit authorization. Rotation, revocation, and provider replacement preserve the Flowstate transcript but end or change future external access.

User-added External Media is supplied by its named third-party provider or origin. Opening a Board initially shows a local placeholder and sends no media request there. The viewer may load it once or remember that exact provider/ origin on the device; the provider then receives the normal network information inherent in a browser request and applies its own practices. Flowstate does not treat Analytics consent as permission to load external Project media and does not server-fetch arbitrary External Media URLs.

Retention

Successful human edits generate automatic Project recovery snapshots, coalesced per editor over two minutes and bounded to 100 automatic snapshots per Project and 30 days. Named recovery points retain their existing Project lifetime. Retention cleanup runs in bounded scheduled batches. Offline queues and Board caches are per account on the device. Signing out clears them; unsynced changes require explicit discard, and users may first download a JSON copy to their device. Such a downloaded copy is managed by the user, outside Flowstate's browser storage.

Normal project content remains until changed or deleted. Trashed projects, nodes, media, and accounts use the documented 30-day recovery window before permanent cleanup. Media Bank also offers an explicitly confirmed Delete unused media operation: Flowstate rechecks that each selected Asset has zero references across active or recoverable content, history, sharing, AI/Bridge/MCP messages, Buttons, Workflow state, and other canonical relationships, then permanently removes its private stored variants and metadata. Deletion is irreversible; a stale, ambiguous, referenced, unauthorized, or storage-failed Asset is preserved and reported instead. Temporary uploads and incomplete resumable uploads expire after 24 hours; completed upload-control records may remain for up to seven days for bounded cleanup. Security tokens and sessions expire according to their purpose. Operational logs and audit records are retained only as needed for security and maintenance; the beta does not currently publish a single universal retention period for every such record.

Your choices

Workflow start forms, step inputs/outputs and owner review decisions are stored with the Project's run records. Human Project-change/task-completion automation stores bounded event input until it is queued; failed starts may remain for owner review/retry for up to 90 days. Completed, failed and cancelled run history has 90-day retention. Active review waits remain until resolved or cancelled. Saved Text and Task Lists follow ordinary Project-content retention. Activity notices appear on an open Board where Activity is available; no email, push provider or new external AI service is involved. Existing Local AI and explicitly connected Bridge rules still apply.

Signed-in users can update profile data, export account data, revoke other sessions, permanently delete server-verified unused owned media, and request account deletion from Settings. Project owners control memberships, public links, AI grants, Bridge connections, integrations, media download policy, and stored-Asset delivery security. Local AI Settings reports device-local model storage and can remove each installed tier without deleting AI Chat history; clearing browser site data also removes local models/preferences/offline state. Browser-local optional analytics consent can be changed through Cookie settings, although no analytics provider is currently configured. Remembered External Media provider/origin choices are separate and can be reset there.

Security and limits

Flowstate uses HTTPS, authenticated sessions, CSRF protection, role checks, private object storage, artifact integrity checks, validation, and bounded maintenance. No service can promise absolute security. Local AI requires compatible WebGPU hardware/browser/storage, can consume substantial memory, power, and disk, and may be slow or inaccurate; its output is not automatically true or authorized. Project mutations remain server-validated and the user remains responsible for reviewing results. Viewer download controls remove the normal download action and reject the dedicated download route; they are not DRM and cannot prevent a viewer from saving content already delivered for inline playback or display.

Policy status

This beta notice documents current behavior and will be updated when the product or its operators' published legal details change. The beta currently does not publish a separate legal-entity name, postal address, or privacy-contact address; none is invented in this notice.